Multimodal RAG Solution for Oil and Gas Drilling Data Processing

infosys · Energy · 2025 · source Agentic: no · promotional bias: high · source-backed · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Oil and gas drilling document intelligence

Infosys needed to extract reliable answers from large volumes of drilling documentation that mix prose, images, charts, logs, and technical diagrams.

Constraints:

Architecture

Infosys built a multimodal RAG system on Amazon Bedrock with hybrid search and parent-child chunking so drilling reports, logs, charts, and lithology diagrams could be retrieved and answered together.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  drilling-document-corpus["Drilling Document Corpus"]
  parent-child-chunking["Parent-Child Chunking"]
  hybrid-search-index["Hybrid Search Index"]
  bedrock-answer-generation["Bedrock Answer Generation"]
  drilling-document-corpus --> parent-child-chunking
  parent-child-chunking --> hybrid-search-index
  hybrid-search-index --> bedrock-answer-generation
Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component Role Interacts with
Drilling Document Corpus Contains well completion reports, drilling logs, lithology diagrams, charts, and images. Parent-Child Chunking
Parent-Child Chunking Preserves document context while exposing retrievable child chunks. Hybrid Search Index
Hybrid Search Index Combines search approaches for drilling-document retrieval. Bedrock Answer Generation
Bedrock Answer Generation Generates answers grounded in retrieved drilling context. —

Data flow

Drilling documents are chunked into parent-child hierarchies, embedded and indexed for hybrid search, retrieved against user questions, and passed through Bedrock-powered generation to return operational answers.

“implemented a hybrid search system with parent-child chunking hierarchy”summary, stated
“process complex oil and gas drilling documentation containing text, images, charts, and technical diagrams.”summary, stated

Why this architecture

Hybrid search plus hierarchical chunking handled both precise technical terms and broader multimodal context, which basic document processing could not capture.

“Through iterative development exploring various chunking strategies, embedding models, and search approaches, the team ultimately implemented a hybrid search system with parent-child chunking hierarchy”summary, stated
“traditional document processing methods struggle to handle effectively.”summary, stated

Models used: Amazon Bedrock · Embedding models · Hybrid search

Scale

Alternatives considered (roads not taken)

Traditional document processing

Rejected because: It struggled with multimodal drilling documents and technical diagrams.

“traditional document processing methods struggle to handle effectively.”summary, stated

Single chunking or search strategy

Rejected because: Iterative development found better performance with hybrid search and parent-child hierarchy.

“exploring various chunking strategies, embedding models, and search approaches”summary, stated

Patterns

rag-retrieval (retrieval)

The system retrieves drilling-document chunks and diagrams before Bedrock-based answering.

“advanced multimodal Retrieval-Augmented Generation (RAG) solution using Amazon Bedrock”summary, stated

eval-driven-development (evaluation)

The team iterated across chunking, embedding, and search choices to improve retrieval accuracy and latency.

“Through iterative development exploring various chunking strategies, embedding models, and search approaches”summary, stated

Tradeoffs

Use parent-child chunking

Chose Hierarchical context-preserving chunks over Flat chunks only.

Rationale: Technical drilling answers need both precise retrieval and broader document context.

“parent-child chunking hierarchy”summary, stated

Use hybrid search

Chose Combined search approaches over A single retrieval method.

Rationale: The team explored alternatives and selected hybrid retrieval for accuracy and latency.

“various chunking strategies, embedding models, and search approaches”summary, implied

Gotchas (what bit them)

Technical diagrams and charts break text-only document pipelines.

Root cause: Drilling documents encode important facts across text, images, charts, and diagrams.

Resolution: Use a multimodal RAG pipeline that processes those artifact types together.

“documentation containing text, images, charts, and technical diagrams.”summary, stated

Chunking choice materially affects retrieval quality.

Root cause: Technical context can be lost when chunks are too flat or disconnected from parent document context.

Resolution: Use a parent-child chunking hierarchy with hybrid search.

“implemented a hybrid search system with parent-child chunking hierarchy”summary, stated

Evaluation

Infosys reports retrieval accuracy, response latency, manual-processing cost reduction, and user time savings.

Metrics: Retrieval accuracy · Response latency · Manual processing cost reduction · Field engineer/geologist time savings

“achieving 92% retrieval accuracy, sub-2-second response times, and delivering significant operational efficiency gains”summary, stated

Outcomes

The system processes thousands of drilling documents across reports, logs, charts, and lithology diagrams.

“40-50% reduction in manual document processing costs and 60% time savings for field engineers and geologists.”summary, stated